DECEMBER 2015 VOLUME 64 NUMBER 12 IEIMAO (ISSN 0018 … · Unsupervised Consensus Clustering of...

DECEMBER 2015 VOLUME 64 NUMBER 12 IEIMAO (ISSN 0018-9456)

EDITORIAL

Editor-in-Chief’s Year-End Message .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. Ferrero 3151

REVIEWERS

2015 List of Reviewers .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3153

REGULAR PAPERS

Acoustics and Ultrasonics Instrumentation and MeasurementsSynchronization Technique for Doppler Signal Extraction in Ultrasonic Vibration Measurement Systems .. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . K. M. Singh and P. Sumathi 3162

A/D and D/A Conversion, Analog and Digital InstrumentationA Low-Power Resistance-to-Frequency Converter Circuit With Wide Frequency Range .. . . . . . . . . . . . . . . K. C. Koay and P. K. Chan 3173ADC Standard IEC 60748-4-3: Precision Measurement of Alternative ENOB Without a Sine Wave . . . . . . . . . . . . . . . . . . R. A. Belcher 3183Accurate Sine-Wave Amplitude Measurements Using Nonlinearly Quantized Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . P. Carbone, J. Schoukens, I. Kollár, and A. Moschitta 3201

Instrumentation for Measurement in CommunicationsHough Transform-Based Clock Skew Measurement Over Network . . . . . . . . . . . . . . . . . . . . K. Oka Saputra, W.-C. Teng, and T.-H. Chen 3209Low-Cost Measurement for a Secondary Mode S Radar Transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Á. Parra-Cerrada,

V. González-Posadas, J. L. Jiménez-Martín, Á. Blanco-del-Campo, W. Hernandez, and C. Calderón-Córdova 3217A Mitigation Technique for Harmonic Downconversion in Wideband Spectrum Sensors . . . . . . . . . . . . . . . . . V. Khatri and G. Banerjee 3226

Controls and Control Systems Instrumentation and MeasurementDynamics Modeling and Measurement of the Microvibrations for a Magnetically Suspended Flywheel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C. Peng, J. Fang, and P. Cui 3239Wave Filtering and State Estimation in Dynamic Positioning of Marine Vessels Using Position Measurement . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . M. Loueipour, M. Keshmiri, M. Danesh, and M. Mojiri 3253

Instrumentation for Measurement of Electric Power Systems, Energy Metering, Electric Power QualityA Robust Technique for Single-Phase Grid Voltage Fundamental and Harmonic Parameter Estimation .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . M. S. Reza and V. G. Agelidis 3262

(Contents Continued on Page 3149)

(Contents Continued from Front Cover)

Compressive Sensing of a Taylor–Fourier Multifrequency Model for Synchrophasor Estimation .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . M. Bertocco, G. Frigo, C. Narduzzi, C. Muscas, and P. A. Pegoraro 3274

Low-Complexity Least-Squares Dynamic Synchrophasor Estimation Based on the Discrete Fourier Transform .. . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D. Belega, D. Fontanelli, and D. Petri 3284

Uncertainty Analysis, Accuracy, Precision and Parameter EstimationUnsupervised Consensus Clustering of Acoustic Emission Time-Series for Robust Damage Sequence Estimation in Composites .. . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E. Ramasso, V. Placet, and M. L. Boubakar 3297Uncertainty Evaluation of a 10 V RMS Sampling Measurement System Using the AC Programmable Josephson Voltage

Standard .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S.-F. Chen, Y. Amagai, M. Maruyama, and N.-h. Kaneko 3308A Modified Nonlinear Two-Filter Smoothing for High-Precision Airborne Integrated GPS and Inertial Navigation .. . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . X. Gong, J. Zhang, and J. Fang 3315

Imaging Techniques and InstrumentationQuantitative Assessment of Flame Stability Through Image Processing and Spectral Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D. Sun, G. Lu, H. Zhou, Y. Yan, and S. Liu 3323Coupling of Fluid Field and Electrostatic Field for Electrical Capacitance Tomography .. . . . . . J. Ye, H. Wang, Y. Li, and W. Yang 3334

Medical and Biomedical Instrumentation and ApplicationsMeasurement of the Heat Removed by Devices for Skin Tags Treatment . . . . . . . . . . . M. Tarabini, B. Saggin, and D. Scaccabarozzi 3354Novel Noncontact Dry Electrode With Adaptive Mechanical Design for Measuring EEG in a Hairy Site .. . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Y.-C. Chen, B.-S. Lin, and J.-S. Pan 3361Wireless Instrumented Crutches for Force and Movement Measurements for Gait Monitoring .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E. Sardini, M. Serpelloni, and M. Lancini 3369

Measurement TechniquesImpedance Comparison Using Unbalanced Bridge With Digital Sine Wave Voltage Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . R. Rybski, J. Kaczmarek, and K. Kontorski 3380Hysteresis Switch Adaptive Velocity Evaluation and High-Resolution Position Subdivision Detection Based on FPGA .. . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . H. Huang and W. Chou 3387Beyond Traditional Clinical Measurements for Screening Fears and Phobias . . . . . . . . . . . . . . . . . P. J. Rosa, F. Esteves, and P. Arriaga 3396Compressed Sensing: A Simple Deterministic Measurement Matrix and a Fast Recovery Algorithm .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. Ravelomanantsoa, H. Rabah, and A. Rouane 3405A Relaxation Oscillator-Based Transformer Ratio Arm Bridge Circuit for Capacitive Humidity Sensor .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . T. Islam, S. A. Khan, M. F. A. Khan, and S. C. Mukhopadhyay 3414Signal Processing Method Based on First-Order Derivative and Multifeature Parameters Combined With Reference Curve for

GWRLG ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . M. Wei, K.-J. Xu, and Z. Liu 3423The Characterization of Pulverized-Coal Pneumatic Transport Using an Array of Intrusive Electrostatic Sensors . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B. Jurjevcic, A. Senegacnik, B. Drobnic, and I. Kuštrin 3434

Metrology, Standards and Calibration

Millimeter-Wave Thermoelectric Power Transfer Standard .. . . . . . . . . R. H. Judaschke, K. Kuhlmann, T. M. Reichel, and W. Perndl 3444Identification and Control of a Cryogenic Current Comparator Using Robust Control Theory .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . M. E. Bierzychudek, R. S. Sánchez-Peña, and A. Tonina 3451Application of a 10 V Programmable Josephson Voltage Standard in Direct Comparison With Conventional Josephson Voltage

Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Y.-H. Tang, J. Wachter, A. Rüfenacht, G. J. FitzPatrick, and S. P. Benz 3458Development of an Electrostatic Calibration System for a Torsional Micronewton Thrust Stand .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . K. H. Cheah, K.-S. Low, Q.-V. Tran, and Z. Lau 3467

Networking, Networks and Sensor NetworksOne-Way Active Delay Measurement With Error Bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . T. Eylen and C. F. Bazlamaçci 3476

Nondestructive Evaluation and Remote SensingCondition Monitoring of Instrumentation Cable Splices Using Kalman Filtering .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S. J. Chang, C. K. Lee, C.-K. Lee, Y. J. Han, M. K. Jung, J. B. Park, and Y.-J. Shin 3490

Optical Instrumentation, Measurement and SystemsElectronic Interface With Vignetting Effect Reduction for a Nikon 6B/6D Autocollimator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . G. J. Bergues, L. Canali, C. Schurrer, and A. G. Flesia 3500Tilted Fiber Bragg Grating Sensors for Reinforcement Corrosion Measurement in Marine Concrete Structure .. . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . M. R. Islam, M. Bagherifaez, M. M. Ali, H. K. Chai, K.-S. Lim, and H. Ahmad 3510

(Contents Continued on Page 3150)

(Contents Continued from 149)

RF, Microwave, Millimeter Wave and Tera-HertzDesign of an FMCW Radar Altimeter for Wide-Range and Low Measurement Error . . . . . . . J.-H. Choi, J.-H. Jang, and J.-E. Roh 3517Application of Electrically Invisible Antennas to the Modulated Scatterer Technique . . . . . . . . . . . . D. A. Crocker and K. M. Donnell 3526On the Use of Magnetically Coupled Resonators for Chirp-Based Timestamping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . A. De Angelis, M. Dionigi, A. Moschitta, P. Carbone, E. Sisinni, P. Ferrari, A. Flammini, and S. Rinaldi 3536Using Gaussian-Uniform Mixture Models for Robust Time-Interval Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. De Angelis, G. De Angelis, and P. Carbone 3545

Sensors, Sensor Fusion and TransducersSelf-Oscillating Fluxgate-Based Quasi-Digital Sensor for DC High-Current Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . N. Wang, Z. Zhang, Z. Li, Y. Zhang, Q. He, B. Han, and Y. Lu 3555Unscented Attitude Estimator Based on Dual Attitude Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . K. Li, L. Chang, and B. Hu 3564UWB-Aided Inertial Motion Capture for Lower Body 3-D Dynamic Activity and Trajectory Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . S. Zihajehzadeh, P. K. Yoon, B.-S. Kang, and E. J. Park 3577

Signals, Systems and System IdentificationWavelet Kernel Local Fisher Discriminant Analysis With Particle Swarm Optimization Algorithm for Bearing Defect

Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . M. Van and H.-J. Kang 3588Wavelet Transform With Histogram-Based Threshold Estimation for Online Partial Discharge Signal Denoising . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . R. Hussein, K. B. Shaban, and A. H. El-Hag 3601Frequency Response Matrix Estimation From Partially Missing Data—for Periodic Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . D. Ugryumova, R. Pintelon, and G. Vandersteen 3615

Time and Frequency Domain Circuits and SystemsImaging Microwave and DC Magnetic Fields in a Vapor-Cell Rb Atomic Clock . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C. Affolderbach, G.-X. Du, T. Bandi, A. Horsley, P. Treutlein, and G. Mileti 3629

INDEXES 3638

IEEE INSTRUMENTATION AND MEASUREMENT SOCIETYThe Instrumentation and Measurement Society is an organization within the framework of the IEEE, of members with principal professional interest in instrumentation and measurement. All IEEE membersare eligible for membership in the Society and will receive this TRANSACTIONS upon payment of the annual Society membership fee of $28.00. For information on joining, write the IEEE at the address below.Member copies of Transactions/Journals are for personal use only.

REZA ZOUGHI,Society President

RUTH A. DYER,Executive VP

DARIO PETRI,,VP Finance

MARK YEARY,VP

Publications

ALESSANDRA FLAMMINI,VP Conferences

SHERVIN SHIRMOHAMMADI,VP Membership

MAX CORTNER,,VP Educations

MIHAELA ALBU, ,VP Technical & Standards

FRANK REYES,,Treasurer

AdCom Term Expires 2015PASQUALE DAPONTEKRISTEN M. DONNELLWENDY VAN MOERJENNY WIRANDI

AdCom Term Expires 2016MIHAELA ALBUALESSANDRA FLAMMINIRUQIANG YANMARK YEARY

AdCom Term Expires 2017LEE BARFORDMAX CORTNERFERDINANDA PONCISHERVIN SHIRMOHAMMADI

AdCom Term Expires 2018SALVATORE BAGLIOZHENG LIUDARIO PETRIJUAN MANUEL RAMIREZ CORTÉS

KIM FOWLER, Society Sr. Past President: JORGE DAHER, Society Jr. Past President:Other Positions on AdCom:THOMAS ROTH, Undergrad Student Rep:MOHAMED KHALIL, Graduate Student Rep:AMY JONES, Young Professionals Rep:SERGIO RAPUANO, Chapter Chair Liason:

ALESSANDRO FERRERO, Transactions Editor-in-Chief:WENDY VAN MOER, Magazine Editor-in-Chief:

MARY WARD-CALLAN, IEEE Staff Director, Technical Activities:JERRY HUDGINS, Division II Director-ElectJUDY SCHARMANN, Society Executive Assistant:CHRIS DYER, Conference Catalysts LLC:

Standing Committee Chairmen, Awards and Recognition Committee:

KIM FOWLER, Chair:KIM FOWLER, Society Awards:ROBERT GOLDBERG, Fellows Evaluation:RUTH DYER, Fellows Coordination:ALESSANDRA FLAMMINI, Conferences and Meetings:DARIO PETRI, FRANK REYES, Finance:

SHERVIN SHIRMOHAMMADI, Membership Development and Services:MAX CORTNER, Education:KRISTEN DONNELL, Distinguished Lecturer Program:JORGE DAHER, Nominations:

MARK YEARY, Publications:ALESSANDRO FERRERO, Transactions Editor-in-Chief:WENDY VAN MOER, Magazine Editor-in-Chief:RUTH A DYER, Society Website:CONFERENCE CATALYSTS, LLC , Webmaster:RUTH DYER, Society Management:MIHAELA ALBU, Technical and Standards:

Technical Committee ChairmenTC-1TC-2TC-3TC-4TC-6TC-7TC-8

TC-9TC-10

TC-13TC-15TC-16TC-17TC-18

ANTHIMOS GEORGIADIS, Measurement Precision, Sensitivity and Noise:YICHENG WANG, DC-LF Measurement:OPEN , Frequency and Time:BRIAN LEE, HF Instrumentation and Measurement:VINCENZO PIURI, Emerging Technologies:LASZLO SUJBERT, RUQIANG YAN, Signals and Systems in Measurements:OPEN , Test and Instrument Systems:OPEN , Instrument/Computer Interface:KANG LEE GEORG BRASSEUR, Sensor Technology:THOMAS LINNENBRINK, Waveform Measurement and Analysis:WILLIAM BOYER, Waveform Recorder:STEVE TILDEN, A/D Converters:STEVE TILDEN, D/A Converters:NICOLAS PAULTER, Pulse Techniques:TRAVIS ELLIS, Probe Standards:SERGIO RAPUANO, Jitter:REINER THOMAE, Wireless & Telecommunications:EMIL PETRIU, Virtual Systems:THIERRY BOSCH, Laser & Optical Systems:JACOB SCHARCANSKI, Materials in Measurement:MICHAEL GARDEnvironmental Measurement:

TC-19TC-20TC-22TC-23TC-24TC-25

TC-26TC-27TC-28TC-30TC-32TC-33

TC-34TC-36TC-37TC-39

GEORGE GIAKOS AND GEORGE ZENTAI , Imaging Systems Measurements:GEORG BRASSEUR AND FRANS GROEN, Transportation:FABIO SCOTTI AND MEL SIEGEL, Intelligent Measurement Systems:OPEN , Education in Instrumentation and Measurements:ANDRZEJ BARWICZ , Measurement Microsystems:MARCO PARVIS, Medical Measurements:VOICU GROZA, Blood Pressure Measurements:MARK YEARY, Radar Cross-Section Measurements:MEL SIEGEL AND PETER WIDE, Human Computer Interface:MEL SIEGEL AND EMIL PETRIU, Robotics & Automation:MEL SIEGEL, VINCENZO PIURI, AND EMIL PETRIU , Security and Contraband Detection:SERGE DEMIDENKO AND NOHPILL PARK, Fault Tolerant Measurement:

MARC VAN DEN BOSCHE AND YVES ROLAIN, Electrical HF and Optical Nonlinear Components:CINDY HARNETT , Nanotechnology in I&M:ZHENG LIU AND DAVID FORSYTH, Industrial Inspection:C. NARDUZZI AND A. EL SADDIK, Measurements for Networking:LORENZO PERETTO AND CARLO MUSCAS, Measurements in Power Systems:

Society Representatives and Directed DelegatesRUTH A DYER, Chair,ROBERT RASSA,

JEFFREY RUBIN, ROBERT GOLDBERG, AUTOTESTCON Board of Directors:VOICU GROZA, Biometrics Council:ROBERT GOLDBERG IEEE Fellows Evaluation:ALESSANDRA FLAMMINI, SERGE DEMIDENKO, MAX CORTNER, JUAN CARLOS MIGUEZ,

LEE BARFORD, SHERVIN SHIRMOHAMMADI, BERNARDO TELLINI, CHI HUNG HWANG, IMTC:

GEORGE XIAO, Journal of Lightwave Technology:AIME LAY EKUAKILLE, Nanotechnology Council:GOURAB SENGUPTA, ALESSANDRA FLAMMINI,

DENIZ GURKAN, SALVATORE BAGLIO, VEDRAN BILAS, SAS Conference:GEORG BRASSEUR, Sensors Council:STEPHEN DYER, JOHN SCHMALZEL, Systems Council:FERDINANDA PONCI Women in Engineering Liaison:

IEEE OfficersHOWARD E. MICHEL, PresidentBARRY L. SHOOP, President-ElectPARVIZ FAMOURI, SecretaryJERRY L. HUDGINS,TreasurerROBERTO DE MARCA,Past President

SAURABH SINHA,Vice President, Educational ActivitiesSHEILA HEMAMI,Vice President, Publication Services and ProductsWAI-CHOONG WONG,Vice President, Member and Geographic ActivitiesBRUCE P. KRAEMER,President, Standards AssociationVINCENZO PIURI,Vice President, Technical ActivitiesJAMES A. JEFFRIES,, President, IEEE-USA

HIROFUMI AKAGI,Director, Division II

IEEE Executive StaffDR. E. JAMES PRENDERGAST,Executive Director & Chief Operating Officer

THOMAS SIEGERT, Business AdministrationELENA GERSTMANN,Corporate ActivitiesDOUGLAS GORHAM, Educational ActivitiesEILEEN M. LACH,General Counsel & Corporate Compliance OfficerSHANNON JOHNSTON, Human ResourcesCHRIS BRANTLEY, IEEE-USA

DONNA HOURICAN, Information Technology (acting)PATRICK MAHONEY, MarketingCECELIA JANKOWSKI, Member and Geographic ActivitiesANTHONY DURNIAK, PublicationsKONSTANTINOS KARACHALIOS, Standards ActivitiesMARY WARD-CALLAN, Technical Activities

IEEE PeriodicalsTransactions/Journals DepartmentFRAN ZAPPULLASenior Director:

DAWN MELLEY, Editorial Director: PETER M. TUOHY, Production Director:MARTIN J. MORAHAN, Managing Editor: SARA T. SCUDDER, Journals Coordinator:

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT (ISSN 0018-9456) is published monthly by The Institute of Electrical and Electronics Engineers, Inc. Responsibility for the contents rests upon the authors and notupon the IEEE, the Society/Council, or its members. IEEE Corporate Office: 3 Park Avenue, 17th Floor, New York, NY 10016-5997. IEEE Operations Center: 445 Hoes Lane, Piscataway, NJ 08854-4141.NJ Telephone: +1 732981 0060. Price/Publication Information: Individual copies: IEEE Members $57.00 (first copy only), nonmembers $113.00 per copy. (Note: Postage and handling charge not included.) Member and nonmember subscription pricesavailable upon request. Copyright and Reprint Permissions: Abstracting is permitted with credit to the source. Libraries are permitted to photocopy for private use of patrons, provided the per-copy fee of $31.00 is paid throughthe Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. For all other copying, reprint, or republication permission, write to Copyrights and Permissions Department, IEEE Publications Administration, 445 HoesLane, Piscataway, NJ 08854-4141. Copyright © 2015 by The Institute of Electrical and Electronics Engineers, Inc. All rights reserved. Periodicals Postage Paid at New York, NY and at additional mailing offices. Postmaster: Sendaddress changes to IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, IEEE, 445 Hoes Lane, Piscataway, NJ 08854-4141. GST Registration No. 125634188. CPC Sales Agreement #40013087. Return undeliverableCanada addresses to: Pitney Bowes IMEX, P.O. Box 4332, Stanton Rd., Toronto, ON M5W 3J4, Canada. IEEE prohibits discrimination, harassment and bullying. For more information visit http://www.ieee.org/nondiscrimination.Printed in U.S.A.

Digital Object Identifier 10.1109/TIM.2015.2497401

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 64, NO. 12, DECEMBER 2015 3209

Hough Transform-Based Clock SkewMeasurement Over Network

Komang Oka Saputra, Wei-Chung Teng, Member, IEEE, and Tsung-Han Chen

Abstract— The accurate clock skew measurement of remotedevices over network connections is crucial to device fingerprint-ing and other related applications. Current approaches use thelower bound of offsets between the target device and the measurerto estimate clock skew; however, the accuracy of estimationis severely affected when even a few offsets appear below thecrowd of offsets. This paper adopted the Hough transform todevelop a new method, which searches for the densest partof the whole distribution. This method is effective in filteringout the upper and lower outliers such that the skew valuesderived from the remaining offsets are stable, even when loweroutliers occur, or when the measuring time is not long enough forcurrent approaches to achieve stable results. The experimentalevaluation of the proposed method has been conducted in orderto compare its performance with that of linear programmingalgorithm (LPA) and two other approaches. During the fiveconsecutive measurements of 1000 offsets each, skews of theproposed method varied within the range of 0.59 ppm, whereasLPA resulted in the range of 0.89 ppm. Both ranges increasedto 1.34 and 63.93 ppm, respectively, when the lower boundsencountered interference from lower outliers.

Index Terms— Clock skew, delay jitter, Hough transform,linear programming, low outlier.

I. INTRODUCTION

CLOCK skew, or the clocking rate difference betweentwo digital clocks, has been widely studied over the

last few decades. Initially, researchers studied the clock skeweffect that caused inaccurate delay measurement between thesender and the receiver. Earlier works [2], [4], [16], [19],[21], [29] focused on eliminating this effect and suggestedmethods for estimating clock skew. Furthermore, two prop-erties of clock skew were identified [17], [19], [21]: thestability over time and the ability to distinguish between anytwo devices. These properties make clock skew a potentialcandidate for physical device fingerprinting and identification.For example, a pioneering work [17] used clock skew asa tool for fingerprinting computers in a general network.Other works [20], [27] utilized the clock skew in revealing

Manuscript received January 26, 2015; revised April 20, 2015; acceptedApril 23, 2015. Date of publication July 21, 2015; date of current versionNovember 6, 2015. This work was supported by the Ministry of Scienceand Technology under Grant NSC-101-2221-E-011-062-MY2. The AssociateEditor coordinating the review process was Dr. Dario Petri.

K. Oka Saputra is with the Department of Computer Science and Infor-mation Engineering, National Taiwan University of Science and Technology,Taipei 10507, Taiwan, and also with the Department of Electrical andComputer Engineering, Udayana University, Bali 80361, Indonesia (e-mail:[email protected]).

W.-C. Teng and T.-H. Chen are with the Department of Computer Scienceand Information Engineering, National Taiwan University of Science andTechnology, Taipei 10507, Taiwan (e-mail: [email protected]).

Digital Object Identifier 10.1109/TIM.2015.2450293

Fig. 1. Offset-set with a stable lower bound depicted by the minimum offsets.

a hidden service behind the onion router (TOR) network.Likewise, few studies [11]–[13], [18], [25] exploited theclock skew to secure time synchronization among sensornodes in a wireless sensor network. Recently, usage of clockskew has been extended as an attack or defense instrumentin various advanced technologies: 1) wireless local areanetwork [3], [15]; 2) cloud environments [14]; 3) mobilehandheld devices [23], [24]; and 4) smartphones [7]. Due tothe importance of clock skew in many areas, it is crucial topursue the accuracy of clock skew measurements.

A clock skew measurement is initialized by collectingtimestamps sent from a device. The measurer can activelysend ICMP requests to the device and collect timestamps fromthe response packets [7], [17], [23], [24]. Alternatively, themeasurer can provide a service (e.g., Web application) bywhich devices communicate with it and send their timestampsthrough AJAX packets [14] or TCP timestamp options inTCP packets [23]. Fig. 1 shows a timestamp collectionsample from an experiment, which will be introduced laterin Section IV-A. In this scatter diagram, each point repre-sents one received timestamp, and the offset of each pointis calculated by subtracting the devices timestamp from themeasurer’s receiving time [14], [19]. After this, a line-fittingmethod such as linear regression can be used to estimate theslope or the clock skew of the whole collection [19], [21].However, variable delays between the device and the mea-surer can make it difficult to determine the slope. Multipathrouting and the accumulated propagation delay, especially in

0018-9456 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

3210 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 64, NO. 12, DECEMBER 2015

wireless communication, might significantly affect the end-to-end delay. The result is the production of many outlier offsetsthat depart from most offsets, located in the lower region ofthe scatter diagram. Consequently, the use of linear regressionalone is not sufficient [14], [19], [21].

It is worth noting that various methods have been deployedto address this problem [2], [4], [14], [15], [16], [19],[21], [29]. For instance, Aoki et al. [2], Huang et al. [14], andPaxson [21] proposed outlier filtering methods by selectingthe minimum offsets of the collected offsets. In the process,Paxson [21] finalized the outlier filter using the median lineprocedure, Aoki et al. [2] used the linear regression to calcu-late the slope of the accumulated minimum offsets selectedfrom several minimum windows, and Huang et al. [14]used the quick piecewise minimum (QPM) algorithm tocalculate the slope only from the minimum offsets in thefirst segment and the last segment of the collected offsets.However, the most widely adopted method is linear program-ming algorithm (LPA), as its result is not significantly affectedby outliers. Moon et al. [19] are the pioneers of clock skewmeasurement by LPA and determine clock skew from thegradient of a line that lies below all the offsets.

All the above methods focused on achieving an accurateclock skew in the presence of outliers. Most of them fullyutilized the characteristics of the offset collection: most offsetsgather in a cluster, outlier offsets happen only above thecluster, and the bottom of the cluster can usually be boundedby a line whose slope is the clock skew. However, this paperobserved that outliers below the cluster, referred to as the lowoutliers, may also occur in network connections with highjitter. In addition, the slopes of the lower bound lines affectedby these low outliers tend to vary in a much larger rangecompared with those in classic cases. Therefore, this paperaims to develop a new estimation method for offset collectionswith low outliers.

The contribution of this paper is twofold. First, it introducesa new method based on the Hough transform: a method thatcombines the concept of clock skew and the Hough transformvoting process. The goal of this method is to identify theparallelogram-like region that encloses the cluster of offsets,enabling the skew estimation to be derived from offsets that areundisrupted by the low outliers as well as the outliers. Second,the proposed method is used to improve the clock skew mea-surement when the lower bound is unstable due to the presenceof the low outliers. The experimental results show that theproposed method always provides stable estimation, comparedwith three other methods: 1) LPA [19]; 2) QPM [14]; and3) Aoki’s method [2], especially in short-time measurement.To conclude, the proposed method is appropriate for allthe existing applications related to clock skew, but it isespecially suitable for applications like device fingerprinting,which requires stable estimation even in wireless yet high-jitternetwork connections.

The following section explains in detail why the lowerbound concept is invalid to offset collections with low outliers.Section III introduces the proposed Hough transform-basedclock skew measurement method. Next, the evaluation resultsand the comparison of the proposed method with existing ones

Fig. 2. Offset-set with an unstable lower bound due to the presence of lowoutliers.

are shown in Section IV. Section V presents several discussionsrelated to the proposed method. The conclusion of this paperis given in Section VI.

II. CLOCK SKEW MEASUREMENT AND THE

USE OF LOWER BOUND

To explain the concept of clock skew, this paper uses the ter-minology of [14] and [19]. For any real-world time t , the localtime reported by the clock of device d is denoted by Cd (t),and its first derivative, or the speed its clock progresses,is denoted by C

′d ≡ dCd (t)/dt , ∀ t ≥ 0. If Cm is

denoted by the clock of the measurer, the skew δ of Cd

can be obtained relative to Cm at time t by calculatingδ(t) = C

′d (t)−C

′m(t). However, in practice, the measurer can

only obtain the offset Cm(t2)−Cd (t1), where t1 is the sendingtime and t2 is the receiving time. This situation is furtheraggravated by the fact that the difference between the sendingtime and the receiving time includes the communication delaysuch that the difference varies for each offset. To reacha stable and confident estimation of skew δ, one possiblemethod is to pick offsets with very close, if not the same,communication delays. As explained in the previous section,it has been observed that offsets with minimal delays in thescatter diagram will most often line up after collecting a fewhundred offsets. It is thus reasonable to assume that delays ofthese offsets are close to the physical minimum delay, and theslope of this straight line is very close to the real skew.

The use of the lower bound itself or its concept can befound in [17], [19], and [22]. As an example, the black circlesin Fig. 1 represent offsets of minimal values in a slidingwindow (the local minimum), and it is obvious that theseblack circles step up slowly but constantly as time goes by.Without being affected by the outliers above, the lower boundline keeps extending in the same slope from the beginning tothe end. Nevertheless, there are cases where the lower boundline breaks up, as shown in Fig. 2 (a sample that will beanalyzed later in Section IV-B). Even though the offsets underthe lower bound line are still close to it, these anomaliesare marked as low outliers. Given a long period, it is still

SAPUTRA et al.: HOUGH TRANSFORM-BASED CLOCK SKEW MEASUREMENT OVER NETWORK 3211

Fig. 3. Relations between ρ, θ , ω, and β in the Hough transform.

possible to obtain a constant lower bound line. For short-timemeasurement, however, parts of the local minimal offset fallbelow the others irregularly, which makes the calculated clockskew fluctuate each time.

It is clear that the existing approaches [2], [14], [19], [21]that utilize the minimum offsets to obtain accurate clockskews become invalid in the context of cases like Fig. 2.It is also a challenging task to filter out the low outliersfrom the concentrated area, because some low outliersare so close to the area, and normally, they should becategorized as part of the cluster of offsets. This paperabandons the use of the lower bound and instead obtainsthe clock skew, the slope of a line, from the concentratedarea itself. As the concentrated area in normal cases likeFigs. 1 and 2 forms the shape of a thick line, it is argued thatthis approach is promising for deriving accurate skews.

III. HOUGH TRANSFORM-BASED CLOCK SKEW

MEASUREMENT

A. Hough Transform

The Hough transform is a commonly used technique toidentify lines from an image [9], [10]. The conventionalHough transform maps the Euclidean coordinate of each dotin the image into another parameter space to quickly find thecandidate lines. Fig. 3 explains the parameters used by theHough transform in finding a straight line from a point set.For point A = (x, y) and a given degree θ , there is exactly oneline that passes through point A, and the angle between thex-axis and its normal line is θ . Here, the perpendicular distancefrom the origin to this line is denoted by ρ. The relationsbetween these variables can thus be summarized as (1). In fact,each (ρ, θ) pair defines one unique candidate line

ρ = x ∗ cos θ + y ∗ sin θ. (1)

To determine the straight lines from many candidates, theHough transform uses a voting mechanism by which each

point votes for all possible lines that pass through it, and thevote results are represented in the form of a matrix indexedby (ρ, θ). The most possible lines are those cells with morevotes than a given threshold [5].

Unlike the conventional Hough transform, the goal of thispaper is to search for a parallelogram-like region instead ofa thin straight line. Thus, the result of this approach willbe θ , a lower bound, and an upper bound of distance ρ, butnot a specific (ρ, θ) pair. One way to denote this regionis (ρl , ρu, θ), where ρl ≤ ρu . Values of ρl and ρu result fromthe clock difference and transmission delay, but the thicknessω = ρu − ρl is basically determined by the jitter. Thus, anew algorithm must be developed to search for the appropriatethickness.

B. Offset Voting for Candidate Regions

Before the new algorithm is designed, the offsets in a scatterdiagram like Fig. 1 must be mapped into image points. All thepoints are mapped to an image plane with nonnegative coor-dinates. This is realized by choosing an appropriate positionin the scatter diagram as the origin of the image. If the offsetsare numbered by the order of their receiving time, then thecoordinate of the i th offset can be expressed as (ti , oi ), where tiis the receiving time of the measurer and oi is the offset value.Furthermore, let omin represent the minimum value of all theoffsets, then the origin chosen is at (t1, omin), and the mappingcan be expressed as (xi , yi ) = (ti − t1, oi − omin).

To find two parallel lines with the same θ accurately,Algorithm 1 was developed based on the voting process ofthe Hough transform in [5]. In Algorithm 1, S stands for theset of all the points; θmin and θmax bound the range of anglefor line detection; p stands for the step size in radians, or thetime precision sought; ωmin is the lower bound of the thicknesssought in the voting process; and ωinc is the increment bywhich the thickness grows. Finally, N stands for the minimumnumber of points a candidate region has to cover. To startthe algorithm, some variables must be declared: L is usedto store all the found candidate regions, and β is the indexof the regions. Given some angle θ and some thickness ω,the whole 2-D space can be divided into many regions byparallel lines with the distance between adjacent lines being ω.Starting with a line of slope θ passing though the origin, theregions can be numbered starting from 0, as shown in Fig. 3.In other words, β is the rounded value if ρ is rounded downto an integer multiple of ω. The Votes(ρ, θ ) used in theHough transform then becomes Votes(β, ω, θ ) accordingly.Algorithm 1 starts by setting the thickness to ωmin. Thefollowing while loop is used to find an eligible region of somethickness in each round. Inside the while loop is a Houghtransform-like process: each point in S votes for all possibleθ between θmin and θmax. If there are one or more angle-region tuples whose votes are at least N , then the tuple withthe highest vote is stored in L, the returned value of functionOffsetVote(). If there is no candidate found, then the thicknessω is increased by ωinc, and a new voting round is started.

It is useful to first try a minimum thickness and graduallyincrease it if necessary, but not the other way around. Whenthe thickness is small, only the area of the highest density


Algorithm 1 Offset Voting Algorithm

function OFFSETVOTE S, θmin , θmax , p, ωmin , ωinc, NV otes = 0L = nullω = ωmin

while L == null dofor all (x, y) in S do

for θ = θmin ; θ ≤ θmax ; θ = θ + p doρ = x * cos θ + y * sin θβ = �ρ / ωV otes(β, θ) = V otes(β, θ) + 1

end forend forfor all (β, θ) do

if V otes(β, θ) ≥ N thenif L == null or

V otes(β, θ) > L .V otes thenL = (β, ω, θ, V otes(β, θ))

end ifend if

end forif L == null then

ω = ω + ωincend if

end whilereturn L

end function

will be chosen, but the enclosed number of offsets may notbe enough to prove the majority. Conversely, a large thicknesswill risk including low outliers and normal outliers in the area.Appropriate minimum thickness and increment values willsave computation time, yet still ensure accurate results. Thispaper used a 500-μs minimum thickness (ωmin) and a100-μs increment (ωinc) in all the experiments.This configuration means that OffsetVote() will start thevoting process from an initial thickness of 500 μs. If nocandidate region is found in the second for loop, the thicknessto try (ω) increases by 100 μs at the end of the while loop.Thus, the value of the final thickness stored in L can beexpressed as 500 + 100k, where k is a nonnegative integer.On the other hand, the minimum number of points N servesas a threshold to ensure the representativeness of the electedregions. If the value of N is too small, OffsteVote() will stopearly with an undesirably small thickness, and the angle ofthe chosen region can obviously differ from the clock skew.The value of N should be at least 35% of the number ofoffsets, even in high-jitter cases. In this paper, 50% was usedas the threshold.

C. Three-Stage Processing

One of the advantages of the Hough transform is that itcan detect lines of all the directions. To save computationtime, however, most applications of the Hough transformuse one degree or larger resolution [9], [10]. In contrast,

Algorithm 2 Three-Stage Process

Require: S, ωmin , ωinc, Np = 10−5

θmin = (π / 2) − 750 ∗ 10−6

θmax = (π / 2) + 750 ∗ 10−6

L = OffsetVote(S, θmin , θmax , p, ωmin , ωinc, N)θmin = L.θ − 5 ∗ 10−6

θmax = L.θ + 5 ∗ 10−6

p = p / 10L = OffsetVote(S, θmin , θmax , p, ωmin , ωinc, N)θmin = L.θ − 5 ∗ 10−7

θmax = L.θ + 5 ∗ 10−7

p = p / 10L = OffsetVote(S, θmin , θmax , p, ωmin , ωinc, N)return (L.β, L.ω, L.θ )

clock skew measurement requires precision of at least ppmlevel [15], [17]. If the parameter p in OffsetVote() is setto 10−6, this function will have to execute voting one milliontimes in the 1° range, given any thickness ω. To keep thecomputation time short, it is vital to limit the range of angleby adopting reasonable θmin and θmax. Of all the reviewedstudies, Jana and Kasera [15] reported the highest clock skewof 1105.69 ppm when they measured several access pointsin their experiments. The second highest number is 750 ppmaccording to [7]. Since a difference of 750 ppm means thattwo clocks will differ by more than 30 min/month, it is reallya rare case. In fact, most of the observed skews fall into therange of −200 to 200 ppm [2]. Incidentally, a measured skewwill become its additive inverse if the target device and themeasurer are switched (skewAB = −skewBA) [3], [12], so itis practical to assume that a clock skew of −750 ppm is aspossible as one of 750 ppm. This paper suggests first choosinga maximum skew value, and then using its additive inverse asthe minimum skew value.

A clock skew value can be viewed as the slope of a straightline, and this slope s can be further expressed as a fraction y/x .On the other hand, if the counter-clockwise angle from thex-axis to this line is a, then s = sin a/cos a = tan a. Since sis a very small number, s can be used as the approximationvalue of a. Finally, θ in function OffsetVote() is the angle ofnormal vector, so the values of θmin and θmax are π/2 addedto the minimum and the maximum clock skew, respectively.If −750 and 750 ppm are used as the minimum and maximumclock skews, respectively, the range to angle will become1.570046–1.571546 rad, or 89.96◦–90.04◦, a much smallerrange than the 0◦–180◦ in the conventional Hough transformvoting process. However, there are still 1501 angles to testif the 1-ppm resolution is used, or 15 001 angles for the0.1-ppm resolution. To solve this problem, this paper designeda three-stage process as shown in Algorithm 2.

Function OffsetVote(), or Algorithm 1, is called once in eachstage. In the first stage, OffsetVote() scans the whole anglerange in 10 ppm units to find the region with the highestvotes and store the result in variable L. The second stage


switches to 1-ppm scan level, but OffsetVote() only scans theangle range of 5 ppm near the angle of region L this time.The third stage is basically a copy of the second stage, butit uses a 0.1-ppm scan level over the range of 0.5 ppm nearthe angle of region L. In this way, the number of angles thatOffsetVote() evaluates sharply decreases from 15 001 to 173.This process succeeds because all the offsets are considered toline up roughly in only one direction, so the region found inthe first stage must have the closest clock skew in the 10-ppmresolution. Stages 2 and 3 simply pursue higher resolutions.The outcome of Algorithm 2 is the index, the thickness,and the angle of the region in the 0.1-ppm precision, or aparallelogram-like region that covers the most offsets.

IV. EVALUATION RESULTS

This paper evaluated the correctness and robustness of theproposed measurement using both normal and low-outlierconditions, as shown in Figs. 1 and 2. For each condition, theresult of the proposed method was compared with three othermethods: 1) LPA [19]; 2) QPM [14]; and 3) Aoki’s method [2].One long-term sample was initially used to generate a reli-able estimation of clock skew, and then this sample setwas cut into several segments of short but equal durations.The short-term skews calculated from these segments revealedhow skew would fluctuate as time went by. The skews ofaccumulated offsets and separate skews were further calculatedto observe how the estimated skew converged.

Both the normal and the low-outlier samples were takenusing the same devices. Therefore, the clock skews in bothsamples should be the same, and the measurement methodswould be expected to produce relatively similar clock skewsfor both samples. An ASUS A46C notebook computer withMicrosoft Windows 7 operating system was used as the targetdevice and another ASUS DUO T9300 notebook runningUbuntu 12.04 operating system as the measurer. In eachmeasurement, the target device sent 5000 timestampsat 200-ms intervals, which took approximately 17 min.

A. Estimation on an Offset-Set Without Low Outlier

The result of long-term measurement by the proposedmethod is detailed in Table I. The final result in the third stage,angle θ = 1.5708382 rad with thickness ω = 500 μs, was usedto create a bounded region as shown in Fig. 4. By applyinglinear regression on only the bounded offsets, an estimatedclock skew of 42.03 ppm was obtained. This result is relativelyclose to those of LPA (42.04 ppm), of QPM (42.16 ppm), andof Aoki’s method (42.06 ppm).

To compare the results of these four methods in short-termmeasurement, Tables II and III summarize the skews of accu-mulated offsets and separate skews of short-term segments,respectively. In Table II, 500 offsets increment were usedto calculate all the skews. As expected, LPA and the pro-posed method contributed stable estimations. It is clear fromthe row Max–Min that the proposed method provides morestable estimation compared with QPM and Aoki’s method.By comparing the clock skew column and the θ−π/2 column,it can be observed that the Hough transform is suitable to

TABLE I

RESULTS OF THE THREE-STAGE PROCESS ON

AN OFFSET-SET WITHOUT LOW OUTLIER

Fig. 4. Enclosed region for an offset-set without low outlier,where θ = 1.5708382 rad and ω = 500 μs.

search for the bounded region, but other analytic tools arestill required, like linear regression, to refine the clock skewestimation. Finally, the skews of accumulated offsets of allthe four methods do not differ more than 0.5 ppm in the casesof 2000 or more offsets.

Table III shows separate skews of the four methods in1000 offset segments. Since all the outliers are excluded atfirst, the proposed method is able to contribute even morestable estimation, on average, than LPA.

B. Estimation on an Offset-Set With Low Outliers

The result of long-term measurement by the proposedmethod is detailed in Table IV. The final result of the thirdstage, angle θ = 1.5708387 rad with thickness ω = 700 μs,was used to create a bounding region as shown in Fig. 5.By applying linear regression on only the bounded offsets,an estimated clock skew of 42.29 ppm was obtained, whichwas very close to the result of the previous measurement.As the offset distribution was not as dense as the previous


TABLE II

SKEWS OF ACCUMULATED OFFSETS ON AN OFFSET-SET

WITHOUT LOW OUTLIER

TABLE III

SEPARATE SKEWS ON AN OFFSET-SET WITHOUT LOW OUTLIER

TABLE IV

RESULTS OF THE THREE-STAGE PROCESS ON

AN OFFSET-SET WITH LOW OUTLIERS

one, a larger ω was necessary to enclose a sufficient numberof offsets.

Similar to Table II, Table V lists the skews of accumulatedoffsets obtained using these four methods. Compared withthe long-term result of the proposed method, LPA gives astable result of 41.91 ppm, Aoki’s method has a somewhatinaccurate estimation of 42.62 ppm, and QPM produces anunacceptable result of 45.62 ppm. Observing how the esti-mated skews converge in Table V, it is surprising that evenLPA is significantly affected by the low outliers until thenumber of offsets accumulates to 2500. The skew difference

Fig. 5. Enclosed region for an offset-set with low outliers,where θ = 1.5708387 rad and ω = 700 μs.

TABLE V

SKEWS OF ACCUMULATED OFFSETS ON

AN OFFSET-SET WITH LOW OUTLIERS

TABLE VI

SEPARATE SKEWS ON AN OFFSET-SET WITH LOW OUTLIERS

between the maximum and the minimum is 15.24 ppm, andthis value is still superior to those of QPM and Aoki’smethod. In contrast, the skews of accumulated offsets by theproposed method do not converge notably because the skewsfluctuate within the range of only 1.38 ppm. It is clear that theproposed method is able to provide much more stable results inshort-term data.

It becomes more obvious that the proposed method hasan advantage over others when the separate short-term skewsin Table VI are compared. Table VI shares the same format


TABLE VII

SOME RESULTS ON MEASURING CLOCK SKEWS OF THREE DIFFERENT DEVICES

TABLE VIII

COMPARISON OF COMPUTATION TIME AND ACCURACY BETWEEN FULL SET RESULTS AND HEAD-AND-TAIL RESULTS

with Table III, but reveals more information. Skews by LPAhappen to vary more significantly than the other methods.It is thus observed that LPA is vulnerable to the cases inwhich low outliers occur only near the middle of the wholemeasuring period. Finally, short-term skews by the proposedmethod fluctuate within the range of only 1.34 ppm, which ismuch smaller than 63.93 ppm by LPA, 47.19 ppm by QPM,and 34.08 ppm by Aoki’s method.

Despite the use of the ASUS notebook, low outliers werealso observed when the skews of other devices were measured.Table VII shows a few results from the ASUS notebook andtwo other devices as a reference. The number of low outliersin Table VII is defined as the number of offsets below thebounding region.

V. COMPUTATIONAL COMPLEXITY

The previous sections demonstrated the use of developedtechniques and evaluated the robustness of the proposedmethod, namely, the three-stage Hough transform-based mea-surement. To analyze the time complexity of this method,some variables must be defined first. Let n denote the numberof measured offsets, and ω1, ω2, and ω3 be the counts ofthickness tried at Stages 1–3, respectively. The first for loopin OffsetVote() is clearly O(n), and the second for loop doesnot take more than O(n) time. Thus, the total computationtime is O((151ω1 + 11ω2 + 11ω3)n), which is still O(n).

In the case of 5000 offsets, it takes 47 s to compute theclock skew by a computer with a 2-GHz RAM and an IntelCore-i7 processor. There are two possible ways to reduce the

computation time. The first is to choose a shorter angle rangein the first stage, like [π /2 − 200×10−6, π /2 + 200×10−6].For the second method, previous studies [17], [19], [21] haverevealed that the clock skew of an offset-set is constant fromthe beginning to the end of the measurement. Thus, only thebeginning and the ending parts of all the offsets may be used toreach very close results. Table VIII compares the computationtime and accuracy between the full set results and their head-and-tail versions. The left-hand side of Table VIII is copiedfrom the part of Table V, and the right-hand side of Table VIIIis derived from the results using only the beginning 500 offsetsand the ending 500 offsets. As expected, the computation timeof the full set results grows in a linear fashion as the numberof offsets increases. However, the time used in head-and-tailversion remains within a small range in all the experiments.On the other hand, the measured skews in both versions arebasically of the same value. The biggest difference betweentwo versions is 0.4 ppm in the 4500 offsets case.

VI. CONCLUSION

This paper proposed a new method for estimating theclock skews of remote devices with a network connection.The advantage of this method is that it provides stable estima-tions, especially when the number of timestamps is limited toa few hundreds. It also reported the existence of low outliers,outliers below the crowd of offsets, which may be causedby high-jitter or other issues. Low outliers make the lowerbound of an offset-set unstable, and thus severely interfere withthe estimations by existing approaches. Since the proposedmethod aims to find a parallelogram-like region that encloses


the densest part of the distribution, it is not affected by theselow outliers. To conclude, the proposed method is the mostrobust way to measure clock skews in only a few minutes.

ACKNOWLEDGMENT

The authors would like to thank L.-C. Cheng for developingthe initial system and for conducting preliminary experimentsin this paper.

REFERENCES

[1] T. E. Abrudan, A. Haghparast, and V. Koivunen, “Time synchroniza-tion and ranging in OFDM systems using time-reversal,” IEEE Trans.Instrum. Meas., vol. 62, no. 12, pp. 3276–3290, Dec. 2013.

[2] M. Aoki, E. Oki, and R. Rojas-Cessa, “Measurement scheme for one-way delay variation with detection and removal of clock skew,” ETRI J.,vol. 32, no. 6, pp. 854–862, Dec. 2010.

[3] C. Arackaparambil, S. Bratus, A. Shubina, and D. Kotz, “On thereliability of wireless fingerprinting using clock skews,” in Proc. 3rdACM Conf. Wireless Netw. Security, 2010, pp. 169–174.

[4] J. Bi, Q. Wu, and Z. Li, “On estimating clock skew for one-waymeasurements,” Comput. Commun., vol. 29, no. 8, pp. 1213–1225,May 2006.

[5] Z.-H. Chen, A. W. Y. Su, and M.-T. Sun, “Resource-efficient FPGAarchitecture and implementation of Hough transform,” IEEE Trans.Very Large Scale Integr. (VLSI) Syst., vol. 20, no. 8, pp. 1419–1428,Aug. 2012.

[6] T. Cooklev, J. C. Eidson, and A. Pakdaman, “An implementation ofIEEE 1588 over IEEE 802.11b for synchronization of wireless localarea network nodes,” IEEE Trans. Instrum. Meas., vol. 56, no. 5,pp. 1632–1639, Oct. 2007.

[7] M. Cristea and B. Groza, “Fingerprinting smartphones remotely viaICMP timestamps,” IEEE Commun. Lett., vol. 17, no. 6, pp. 1081–1083,Jun. 2013.

[8] C. M. De Dominicis, P. Pivato, P. Ferrari, D. Macii, E. Sisinni,and A. Flammini, “Timestamping of IEEE 802.15.4a CSS signals forwireless ranging and time synchronization,” IEEE Trans. Instrum. Meas.,vol. 62, no. 8, pp. 2286–2296, Aug. 2013.

[9] R. O. Duda and P. E. Hart, “Use of the Hough transformation to detectlines and curves in pictures,” Commun. ACM, vol. 15, no. 1, pp. 11–15,Jan. 1972.

[10] P. E. Hart, “How the Hough transform was invented [DSP History],”IEEE Signal Process. Mag., vol. 26, no. 6, pp. 18–22, Nov. 2009.

[11] D.-J. Huang, W.-C. Teng, C.-Y. Wang, H.-Y. Huang, andJ. M. Hellerstein, “Clock skew based node identification inwireless sensor networks,” in Proc. IEEE Global Telecommun.Conf. (GLOBECOM), Nov./Dec. 2008, pp. 1–5.

[12] D.-J. Huang and W.-C. Teng, “A defense against clock skew replicationattacks in wireless sensor networks,” J. Netw. Comput. Appl., vol. 39,pp. 26–37, Mar. 2014.

[13] D.-J. Huang, W.-C. Teng, and K.-T. Yang, “Secured flooding timesynchronization protocol with moderator,” Int. J. Commun. Syst., vol. 26,no. 9, pp. 1092–1115, 2013.

[14] D.-J. Huang, K.-T. Yang, C.-C. Ni, W.-C. Teng, T.-R. Hsiang, andY.-J. Lee, “Clock skew based client device identification in cloud envi-ronments,” in Proc. 26th IEEE Int. Conf. Adv. Inf. Netw. Appl. (AINA),Mar. 2012, pp. 526–533.

[15] S. Jana and S. K. Kasera, “On fast and accurate detection of unauthorizedwireless access points using clock skews,” IEEE Trans. Mobile Comput.,vol. 9, no. 3, pp. 449–462, Mar. 2010.

[16] H. Khlifi and J.-C. Grégoire, “Low-complexity offline and onlineclock skew estimation and removal,” Comput. Netw., vol. 50, no. 11,pp. 1872–1884, Aug. 2006.

[17] T. Kohno, A. Broido, and K. C. Claffy, “Remote physical devicefingerprinting,” IEEE Trans. Dependable Secure Comput., vol. 2, no. 2,pp. 93–108, Apr./Jun. 2005.

[18] X. Mei, D. Liu, K. Sun, and D. Xu, “On feasibility of fingerprintingwireless sensor nodes using physical properties,” in Proc. 27th IEEE Int.Symp. Parallel Distrib. Process. (IPDPS), May 2013, pp. 1112–1121.

[19] S. B. Moon, P. Skelly, and D. Towsley, “Estimation and removal ofclock skew from network delay measurements,” in Proc. INFOCOMConf., Mar. 1999, pp. 227–234.

[20] S. J. Murdoch, “Hot or not: Revealing hidden services by their clockskew,” in Proc. 13th ACM Conf. Comput. Commun. Security, 2006,pp. 27–36.

[21] V. Paxson, “On calibrating measurements of packet transit times,” inProc. ACM SIGMETRICS Conf., 1998, pp. 11–21.

[22] L. Polcák, J. Jirásek, and P. Matousek, “Comment on ‘remote physi-cal device fingerprinting,”’ IEEE Trans. Dependable Secure Comput.,vol. 11, no. 5, pp. 494–496, Sep./Oct. 2014.

[23] S. Sharma, H. Saran, and S. Bansal, “An empirical study of clock skewbehavior in modern mobile and hand-held devices,” in Proc. 3rd Int.Conf. Commun. Syst. Netw. (COMSNETS), Jan. 2011, pp. 1–4.

[24] S. Sharma, A. Hussain, and H. Saran, “Experience with heterogenousclock-skew based device fingerprinting,” in Proc. LASER Workshop,2012, pp. 9–18.

[25] M. B. Uddin and C. Castelluccia, “Toward clock skew based wirelesssensor node services,” in Proc. 5th Annu. ICST Wireless InternetConf. (WICON), 2010, pp. 1–9.

[26] A. Vakili and J.-C. Grégoire, “Accurate one-way delay estimation:Limitations and improvements,” IEEE Trans. Instrum. Meas., vol. 61,no. 9, pp. 2428–2435, Sep. 2012.

[27] S. Zander and S. J. Murdoch, “An improved clock-skew measurementtechnique for revealing hidden services,” in Proc. 17th Conf. SecuritySymp., 2008, pp. 211–225.

[28] K. Zeng, K. Govindan, and P. Mohapatra, “Non-cryptographic authen-tication and identification in wireless networks [Security and Privacy inEmerging Wireless Networks],” IEEE Wireless Commun., vol. 17, no. 5,pp. 56–62, Oct. 2010.

[29] L. Zhang, Z. Liu, and C. H. Xia, “Clock synchronization algorithms fornetwork measurements,” in Proc. INFOCOM Conf., 2002, pp. 160–169.

Komang Oka Saputra received the B.Eng. degreein electrical engineering from Brawijaya University,Malang, Indonesia, in 2004, and the M.Eng.degree in electrical engineering from the Univer-sity of Indonesia, Depok, Indonesia, in 2006, witha specialization in telecommunication engineering.He is currently pursuing the Ph.D. degree with theDepartment of Computer Science and InformationEngineering, National Taiwan University of Scienceand Technology, Taipei, Taiwan.

He has been a Faculty Member with the Depart-ment of Electrical and Computer Engineering, Udayana University, Bali,Indonesia, since 2008. His current research interests include computernetwork, networking protocol, and security issues of communication systems.

Wei-Chung Teng (M’09) received theD.Eng. degree from the University of Tokyo,Tokyo, Japan, in 2001.

He is currently an Associate Professor with theDepartment of Computer Science and InformationEngineering, National Taiwan University of Scienceand Technology, Taipei, Taiwan. His current researchinterests include human–computer interactionfocusing on remote robot manipulation, networkcommunication protocols of time synchronization,and network security issues.

Tsung-Han Chen received the B.S. degree from theDepartment of Computer Science and Engineering,National Taiwan Ocean University, Keelung, Taiwan,in 2012, and the M.S. degree from the Departmentof Computer Science and Information Engineering,National Taiwan University of Science andTechnology, Taipei, Taiwan.

His current research interests include softwareengineering and network communicationapplications.

DECEMBER 2015 VOLUME 64 NUMBER 12 IEIMAO (ISSN 0018 … · Unsupervised Consensus Clustering of...

Documents

Transcript of DECEMBER 2015 VOLUME 64 NUMBER 12 IEIMAO (ISSN 0018 … · Unsupervised Consensus Clustering of...